drivers/nvme/host/fabrics.c

   1 /*
   2  * NVMe over Fabrics common host code.
   3  * Copyright (c) 2015-2016 HGST, a Western Digital Company.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms and conditions of the GNU General Public License,
   7  * version 2, as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
  12  * more details.
  13  */
  14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  15 #include <linux/init.h>
  16 #include <linux/miscdevice.h>
  17 #include <linux/module.h>
  18 #include <linux/mutex.h>
  19 #include <linux/parser.h>
  20 #include <linux/seq_file.h>
  21 #include "nvme.h"
  22 #include "fabrics.h"
  23
  24 static LIST_HEAD(nvmf_transports);
  25 static DECLARE_RWSEM(nvmf_transports_rwsem);
  26
  27 static LIST_HEAD(nvmf_hosts);
  28 static DEFINE_MUTEX(nvmf_hosts_mutex);
  29
  30 static struct nvmf_host *nvmf_default_host;
  31
  32 static struct nvmf_host *__nvmf_host_find(const char *hostnqn)
  33 {
  34         struct nvmf_host *host;
  35
  36         list_for_each_entry(host, &nvmf_hosts, list) {
  37                 if (!strcmp(host->nqn, hostnqn))
  38                         return host;
  39         }
  40
  41         return NULL;
  42 }
  43
  44 static struct nvmf_host *nvmf_host_add(const char *hostnqn)
  45 {
  46         struct nvmf_host *host;
  47
  48         mutex_lock(&nvmf_hosts_mutex);
  49         host = __nvmf_host_find(hostnqn);
  50         if (host) {
  51                 kref_get(&host->ref);
  52                 goto out_unlock;
  53         }
  54
  55         host = kmalloc(sizeof(*host), GFP_KERNEL);
  56         if (!host)
  57                 goto out_unlock;
  58
  59         kref_init(&host->ref);
  60         strlcpy(host->nqn, hostnqn, NVMF_NQN_SIZE);
  61
  62         list_add_tail(&host->list, &nvmf_hosts);
  63 out_unlock:
  64         mutex_unlock(&nvmf_hosts_mutex);
  65         return host;
  66 }
  67
  68 static struct nvmf_host *nvmf_host_default(void)
  69 {
  70         struct nvmf_host *host;
  71
  72         host = kmalloc(sizeof(*host), GFP_KERNEL);
  73         if (!host)
  74                 return NULL;
  75
  76         kref_init(&host->ref);
  77         uuid_gen(&host->id);
  78         snprintf(host->nqn, NVMF_NQN_SIZE,
  79                 "nqn.2014-08.org.nvmexpress:uuid:%pUb", &host->id);
  80
  81         mutex_lock(&nvmf_hosts_mutex);
  82         list_add_tail(&host->list, &nvmf_hosts);
  83         mutex_unlock(&nvmf_hosts_mutex);
  84
  85         return host;
  86 }
  87
  88 static void nvmf_host_destroy(struct kref *ref)
  89 {
  90         struct nvmf_host *host = container_of(ref, struct nvmf_host, ref);
  91
  92         mutex_lock(&nvmf_hosts_mutex);
  93         list_del(&host->list);
  94         mutex_unlock(&nvmf_hosts_mutex);
  95
  96         kfree(host);
  97 }
  98
  99 static void nvmf_host_put(struct nvmf_host *host)
 100 {
 101         if (host)
 102                 kref_put(&host->ref, nvmf_host_destroy);
 103 }
 104
 105 /**
 106  * nvmf_get_address() -  Get address/port
 107  * @ctrl:       Host NVMe controller instance which we got the address
 108  * @buf:        OUTPUT parameter that will contain the address/port
 109  * @size:       buffer size
 110  */
 111 int nvmf_get_address(struct nvme_ctrl *ctrl, char *buf, int size)
 112 {
 113         int len = 0;
 114
 115         if (ctrl->opts->mask & NVMF_OPT_TRADDR)
 116                 len += snprintf(buf, size, "traddr=%s", ctrl->opts->traddr);
 117         if (ctrl->opts->mask & NVMF_OPT_TRSVCID)
 118                 len += snprintf(buf + len, size - len, "%strsvcid=%s",
 119                                 (len) ? "," : "", ctrl->opts->trsvcid);
 120         if (ctrl->opts->mask & NVMF_OPT_HOST_TRADDR)
 121                 len += snprintf(buf + len, size - len, "%shost_traddr=%s",
 122                                 (len) ? "," : "", ctrl->opts->host_traddr);
 123         len += snprintf(buf + len, size - len, "\n");
 124
 125         return len;
 126 }
 127 EXPORT_SYMBOL_GPL(nvmf_get_address);
 128
 129 /**
 130  * nvmf_reg_read32() -  NVMe Fabrics "Property Get" API function.
 131  * @ctrl:       Host NVMe controller instance maintaining the admin
 132  *              queue used to submit the property read command to
 133  *              the allocated NVMe controller resource on the target system.
 134  * @off:        Starting offset value of the targeted property
 135  *              register (see the fabrics section of the NVMe standard).
 136  * @val:        OUTPUT parameter that will contain the value of
 137  *              the property after a successful read.
 138  *
 139  * Used by the host system to retrieve a 32-bit capsule property value
 140  * from an NVMe controller on the target system.
 141  *
 142  * ("Capsule property" is an "PCIe register concept" applied to the
 143  * NVMe fabrics space.)
 144  *
 145  * Return:
 146  *      0: successful read
 147  *      > 0: NVMe error status code
 148  *      < 0: Linux errno error code
 149  */
 150 int nvmf_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val)
 151 {
 152         struct nvme_command cmd;
 153         union nvme_result res;
 154         int ret;
 155
 156         memset(&cmd, 0, sizeof(cmd));
 157         cmd.prop_get.opcode = nvme_fabrics_command;
 158         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
 159         cmd.prop_get.offset = cpu_to_le32(off);
 160
 161         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
 162                         NVME_QID_ANY, 0, 0);
 163
 164         if (ret >= 0)
 165                 *val = le64_to_cpu(res.u64);
 166         if (unlikely(ret != 0))
 167                 dev_err(ctrl->device,
 168                         "Property Get error: %d, offset %#x\n",
 169                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
 170
 171         return ret;
 172 }
 173 EXPORT_SYMBOL_GPL(nvmf_reg_read32);
 174
 175 /**
 176  * nvmf_reg_read64() -  NVMe Fabrics "Property Get" API function.
 177  * @ctrl:       Host NVMe controller instance maintaining the admin
 178  *              queue used to submit the property read command to
 179  *              the allocated controller resource on the target system.
 180  * @off:        Starting offset value of the targeted property
 181  *              register (see the fabrics section of the NVMe standard).
 182  * @val:        OUTPUT parameter that will contain the value of
 183  *              the property after a successful read.
 184  *
 185  * Used by the host system to retrieve a 64-bit capsule property value
 186  * from an NVMe controller on the target system.
 187  *
 188  * ("Capsule property" is an "PCIe register concept" applied to the
 189  * NVMe fabrics space.)
 190  *
 191  * Return:
 192  *      0: successful read
 193  *      > 0: NVMe error status code
 194  *      < 0: Linux errno error code
 195  */
 196 int nvmf_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val)
 197 {
 198         struct nvme_command cmd;
 199         union nvme_result res;
 200         int ret;
 201
 202         memset(&cmd, 0, sizeof(cmd));
 203         cmd.prop_get.opcode = nvme_fabrics_command;
 204         cmd.prop_get.fctype = nvme_fabrics_type_property_get;
 205         cmd.prop_get.attrib = 1;
 206         cmd.prop_get.offset = cpu_to_le32(off);
 207
 208         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res, NULL, 0, 0,
 209                         NVME_QID_ANY, 0, 0);
 210
 211         if (ret >= 0)
 212                 *val = le64_to_cpu(res.u64);
 213         if (unlikely(ret != 0))
 214                 dev_err(ctrl->device,
 215                         "Property Get error: %d, offset %#x\n",
 216                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
 217         return ret;
 218 }
 219 EXPORT_SYMBOL_GPL(nvmf_reg_read64);
 220
 221 /**
 222  * nvmf_reg_write32() -  NVMe Fabrics "Property Write" API function.
 223  * @ctrl:       Host NVMe controller instance maintaining the admin
 224  *              queue used to submit the property read command to
 225  *              the allocated NVMe controller resource on the target system.
 226  * @off:        Starting offset value of the targeted property
 227  *              register (see the fabrics section of the NVMe standard).
 228  * @val:        Input parameter that contains the value to be
 229  *              written to the property.
 230  *
 231  * Used by the NVMe host system to write a 32-bit capsule property value
 232  * to an NVMe controller on the target system.
 233  *
 234  * ("Capsule property" is an "PCIe register concept" applied to the
 235  * NVMe fabrics space.)
 236  *
 237  * Return:
 238  *      0: successful write
 239  *      > 0: NVMe error status code
 240  *      < 0: Linux errno error code
 241  */
 242 int nvmf_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val)
 243 {
 244         struct nvme_command cmd;
 245         int ret;
 246
 247         memset(&cmd, 0, sizeof(cmd));
 248         cmd.prop_set.opcode = nvme_fabrics_command;
 249         cmd.prop_set.fctype = nvme_fabrics_type_property_set;
 250         cmd.prop_set.attrib = 0;
 251         cmd.prop_set.offset = cpu_to_le32(off);
 252         cmd.prop_set.value = cpu_to_le64(val);
 253
 254         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, NULL, NULL, 0, 0,
 255                         NVME_QID_ANY, 0, 0);
 256         if (unlikely(ret))
 257                 dev_err(ctrl->device,
 258                         "Property Set error: %d, offset %#x\n",
 259                         ret > 0 ? ret & ~NVME_SC_DNR : ret, off);
 260         return ret;
 261 }
 262 EXPORT_SYMBOL_GPL(nvmf_reg_write32);
 263
 264 /**
 265  * nvmf_log_connect_error() - Error-parsing-diagnostic print
 266  * out function for connect() errors.
 267  *
 268  * @ctrl: the specific /dev/nvmeX device that had the error.
 269  *
 270  * @errval: Error code to be decoded in a more human-friendly
 271  *          printout.
 272  *
 273  * @offset: For use with the NVMe error code NVME_SC_CONNECT_INVALID_PARAM.
 274  *
 275  * @cmd: This is the SQE portion of a submission capsule.
 276  *
 277  * @data: This is the "Data" portion of a submission capsule.
 278  */
 279 static void nvmf_log_connect_error(struct nvme_ctrl *ctrl,
 280                 int errval, int offset, struct nvme_command *cmd,
 281                 struct nvmf_connect_data *data)
 282 {
 283         int err_sctype = errval & (~NVME_SC_DNR);
 284
 285         switch (err_sctype) {
 286
 287         case (NVME_SC_CONNECT_INVALID_PARAM):
 288                 if (offset >> 16) {
 289                         char *inv_data = "Connect Invalid Data Parameter";
 290
 291                         switch (offset & 0xffff) {
 292                         case (offsetof(struct nvmf_connect_data, cntlid)):
 293                                 dev_err(ctrl->device,
 294                                         "%s, cntlid: %d\n",
 295                                         inv_data, data->cntlid);
 296                                 break;
 297                         case (offsetof(struct nvmf_connect_data, hostnqn)):
 298                                 dev_err(ctrl->device,
 299                                         "%s, hostnqn \"%s\"\n",
 300                                         inv_data, data->hostnqn);
 301                                 break;
 302                         case (offsetof(struct nvmf_connect_data, subsysnqn)):
 303                                 dev_err(ctrl->device,
 304                                         "%s, subsysnqn \"%s\"\n",
 305                                         inv_data, data->subsysnqn);
 306                                 break;
 307                         default:
 308                                 dev_err(ctrl->device,
 309                                         "%s, starting byte offset: %d\n",
 310                                        inv_data, offset & 0xffff);
 311                                 break;
 312                         }
 313                 } else {
 314                         char *inv_sqe = "Connect Invalid SQE Parameter";
 315
 316                         switch (offset) {
 317                         case (offsetof(struct nvmf_connect_command, qid)):
 318                                 dev_err(ctrl->device,
 319                                        "%s, qid %d\n",
 320                                         inv_sqe, cmd->connect.qid);
 321                                 break;
 322                         default:
 323                                 dev_err(ctrl->device,
 324                                         "%s, starting byte offset: %d\n",
 325                                         inv_sqe, offset);
 326                         }
 327                 }
 328                 break;
 329
 330         case NVME_SC_CONNECT_INVALID_HOST:
 331                 dev_err(ctrl->device,
 332                         "Connect for subsystem %s is not allowed, hostnqn: %s\n",
 333                         data->subsysnqn, data->hostnqn);
 334                 break;
 335
 336         case NVME_SC_CONNECT_CTRL_BUSY:
 337                 dev_err(ctrl->device,
 338                         "Connect command failed: controller is busy or not available\n");
 339                 break;
 340
 341         case NVME_SC_CONNECT_FORMAT:
 342                 dev_err(ctrl->device,
 343                         "Connect incompatible format: %d",
 344                         cmd->connect.recfmt);
 345                 break;
 346
 347         default:
 348                 dev_err(ctrl->device,
 349                         "Connect command failed, error wo/DNR bit: %d\n",
 350                         err_sctype);
 351                 break;
 352         } /* switch (err_sctype) */
 353 }
 354
 355 /**
 356  * nvmf_connect_admin_queue() - NVMe Fabrics Admin Queue "Connect"
 357  *                              API function.
 358  * @ctrl:       Host nvme controller instance used to request
 359  *              a new NVMe controller allocation on the target
 360  *              system and  establish an NVMe Admin connection to
 361  *              that controller.
 362  *
 363  * This function enables an NVMe host device to request a new allocation of
 364  * an NVMe controller resource on a target system as well establish a
 365  * fabrics-protocol connection of the NVMe Admin queue between the
 366  * host system device and the allocated NVMe controller on the
 367  * target system via a NVMe Fabrics "Connect" command.
 368  *
 369  * Return:
 370  *      0: success
 371  *      > 0: NVMe error status code
 372  *      < 0: Linux errno error code
 373  *
 374  */
 375 int nvmf_connect_admin_queue(struct nvme_ctrl *ctrl)
 376 {
 377         struct nvme_command cmd;
 378         union nvme_result res;
 379         struct nvmf_connect_data *data;
 380         int ret;
 381
 382         memset(&cmd, 0, sizeof(cmd));
 383         cmd.connect.opcode = nvme_fabrics_command;
 384         cmd.connect.fctype = nvme_fabrics_type_connect;
 385         cmd.connect.qid = 0;
 386         cmd.connect.sqsize = cpu_to_le16(NVME_AQ_DEPTH - 1);
 387
 388         /*
 389          * Set keep-alive timeout in seconds granularity (ms * 1000)
 390          * and add a grace period for controller kato enforcement
 391          */
 392         cmd.connect.kato = ctrl->opts->discovery_nqn ? 0 :
 393                 cpu_to_le32((ctrl->kato + NVME_KATO_GRACE) * 1000);
 394
 395         data = kzalloc(sizeof(*data), GFP_KERNEL);
 396         if (!data)
 397                 return -ENOMEM;
 398
 399         uuid_copy(&data->hostid, &ctrl->opts->host->id);
 400         data->cntlid = cpu_to_le16(0xffff);
 401         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
 402         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
 403
 404         ret = __nvme_submit_sync_cmd(ctrl->admin_q, &cmd, &res,
 405                         data, sizeof(*data), 0, NVME_QID_ANY, 1,
 406                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
 407         if (ret) {
 408                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
 409                                        &cmd, data);
 410                 goto out_free_data;
 411         }
 412
 413         ctrl->cntlid = le16_to_cpu(res.u16);
 414
 415 out_free_data:
 416         kfree(data);
 417         return ret;
 418 }
 419 EXPORT_SYMBOL_GPL(nvmf_connect_admin_queue);
 420
 421 /**
 422  * nvmf_connect_io_queue() - NVMe Fabrics I/O Queue "Connect"
 423  *                           API function.
 424  * @ctrl:       Host nvme controller instance used to establish an
 425  *              NVMe I/O queue connection to the already allocated NVMe
 426  *              controller on the target system.
 427  * @qid:        NVMe I/O queue number for the new I/O connection between
 428  *              host and target (note qid == 0 is illegal as this is
 429  *              the Admin queue, per NVMe standard).
 430  *
 431  * This function issues a fabrics-protocol connection
 432  * of a NVMe I/O queue (via NVMe Fabrics "Connect" command)
 433  * between the host system device and the allocated NVMe controller
 434  * on the target system.
 435  *
 436  * Return:
 437  *      0: success
 438  *      > 0: NVMe error status code
 439  *      < 0: Linux errno error code
 440  */
 441 int nvmf_connect_io_queue(struct nvme_ctrl *ctrl, u16 qid)
 442 {
 443         struct nvme_command cmd;
 444         struct nvmf_connect_data *data;
 445         union nvme_result res;
 446         int ret;
 447
 448         memset(&cmd, 0, sizeof(cmd));
 449         cmd.connect.opcode = nvme_fabrics_command;
 450         cmd.connect.fctype = nvme_fabrics_type_connect;
 451         cmd.connect.qid = cpu_to_le16(qid);
 452         cmd.connect.sqsize = cpu_to_le16(ctrl->sqsize);
 453
 454         data = kzalloc(sizeof(*data), GFP_KERNEL);
 455         if (!data)
 456                 return -ENOMEM;
 457
 458         uuid_copy(&data->hostid, &ctrl->opts->host->id);
 459         data->cntlid = cpu_to_le16(ctrl->cntlid);
 460         strncpy(data->subsysnqn, ctrl->opts->subsysnqn, NVMF_NQN_SIZE);
 461         strncpy(data->hostnqn, ctrl->opts->host->nqn, NVMF_NQN_SIZE);
 462
 463         ret = __nvme_submit_sync_cmd(ctrl->connect_q, &cmd, &res,
 464                         data, sizeof(*data), 0, qid, 1,
 465                         BLK_MQ_REQ_RESERVED | BLK_MQ_REQ_NOWAIT);
 466         if (ret) {
 467                 nvmf_log_connect_error(ctrl, ret, le32_to_cpu(res.u32),
 468                                        &cmd, data);
 469         }
 470         kfree(data);
 471         return ret;
 472 }
 473 EXPORT_SYMBOL_GPL(nvmf_connect_io_queue);
 474
 475 bool nvmf_should_reconnect(struct nvme_ctrl *ctrl)
 476 {
 477         if (ctrl->opts->max_reconnects == -1 ||
 478             ctrl->nr_reconnects < ctrl->opts->max_reconnects)
 479                 return true;
 480
 481         return false;
 482 }
 483 EXPORT_SYMBOL_GPL(nvmf_should_reconnect);
 484
 485 /**
 486  * nvmf_register_transport() - NVMe Fabrics Library registration function.
 487  * @ops:        Transport ops instance to be registered to the
 488  *              common fabrics library.
 489  *
 490  * API function that registers the type of specific transport fabric
 491  * being implemented to the common NVMe fabrics library. Part of
 492  * the overall init sequence of starting up a fabrics driver.
 493  */
 494 int nvmf_register_transport(struct nvmf_transport_ops *ops)
 495 {
 496         if (!ops->create_ctrl)
 497                 return -EINVAL;
 498
 499         down_write(&nvmf_transports_rwsem);
 500         list_add_tail(&ops->entry, &nvmf_transports);
 501         up_write(&nvmf_transports_rwsem);
 502
 503         return 0;
 504 }
 505 EXPORT_SYMBOL_GPL(nvmf_register_transport);
 506
 507 /**
 508  * nvmf_unregister_transport() - NVMe Fabrics Library unregistration function.
 509  * @ops:        Transport ops instance to be unregistered from the
 510  *              common fabrics library.
 511  *
 512  * Fabrics API function that unregisters the type of specific transport
 513  * fabric being implemented from the common NVMe fabrics library.
 514  * Part of the overall exit sequence of unloading the implemented driver.
 515  */
 516 void nvmf_unregister_transport(struct nvmf_transport_ops *ops)
 517 {
 518         down_write(&nvmf_transports_rwsem);
 519         list_del(&ops->entry);
 520         up_write(&nvmf_transports_rwsem);
 521 }
 522 EXPORT_SYMBOL_GPL(nvmf_unregister_transport);
 523
 524 static struct nvmf_transport_ops *nvmf_lookup_transport(
 525                 struct nvmf_ctrl_options *opts)
 526 {
 527         struct nvmf_transport_ops *ops;
 528
 529         lockdep_assert_held(&nvmf_transports_rwsem);
 530
 531         list_for_each_entry(ops, &nvmf_transports, entry) {
 532                 if (strcmp(ops->name, opts->transport) == 0)
 533                         return ops;
 534         }
 535
 536         return NULL;
 537 }
 538
 539 /*
 540  * For something we're not in a state to send to the device the default action
 541  * is to busy it and retry it after the controller state is recovered.  However,
 542  * if the controller is deleting or if anything is marked for failfast or
 543  * nvme multipath it is immediately failed.
 544  *
 545  * Note: commands used to initialize the controller will be marked for failfast.
 546  * Note: nvme cli/ioctl commands are marked for failfast.
 547  */
 548 blk_status_t nvmf_fail_nonready_command(struct nvme_ctrl *ctrl,
 549                 struct request *rq)
 550 {
 551         if (ctrl->state != NVME_CTRL_DELETING &&
 552             ctrl->state != NVME_CTRL_DEAD &&
 553             !blk_noretry_request(rq) && !(rq->cmd_flags & REQ_NVME_MPATH))
 554                 return BLK_STS_RESOURCE;
 555
 556         nvme_req(rq)->status = NVME_SC_HOST_PATH_ERROR;
 557         blk_mq_start_request(rq);
 558         nvme_complete_rq(rq);
 559         return BLK_STS_OK;
 560 }
 561 EXPORT_SYMBOL_GPL(nvmf_fail_nonready_command);
 562
 563 bool __nvmf_check_ready(struct nvme_ctrl *ctrl, struct request *rq,
 564                 bool queue_live)
 565 {
 566         struct nvme_request *req = nvme_req(rq);
 567
 568         /*
 569          * If we are in some state of setup or teardown only allow
 570          * internally generated commands.
 571          */
 572         if (!blk_rq_is_passthrough(rq) || (req->flags & NVME_REQ_USERCMD))
 573                 return false;
 574
 575         /*
 576          * Only allow commands on a live queue, except for the connect command,
 577          * which is require to set the queue live in the appropinquate states.
 578          */
 579         switch (ctrl->state) {
 580         case NVME_CTRL_CONNECTING:
 581                 if (req->cmd->common.opcode == nvme_fabrics_command &&
 582                     req->cmd->fabrics.fctype == nvme_fabrics_type_connect)
 583                         return true;
 584                 break;
 585         default:
 586                 break;
 587         case NVME_CTRL_DEAD:
 588                 return false;
 589         }
 590
 591         return queue_live;
 592 }
 593 EXPORT_SYMBOL_GPL(__nvmf_check_ready);
 594
 595 static const match_table_t opt_tokens = {
 596         { NVMF_OPT_TRANSPORT,           "transport=%s"          },
 597         { NVMF_OPT_TRADDR,              "traddr=%s"             },
 598         { NVMF_OPT_TRSVCID,             "trsvcid=%s"            },
 599         { NVMF_OPT_NQN,                 "nqn=%s"                },
 600         { NVMF_OPT_QUEUE_SIZE,          "queue_size=%d"         },
 601         { NVMF_OPT_NR_IO_QUEUES,        "nr_io_queues=%d"       },
 602         { NVMF_OPT_RECONNECT_DELAY,     "reconnect_delay=%d"    },
 603         { NVMF_OPT_CTRL_LOSS_TMO,       "ctrl_loss_tmo=%d"      },
 604         { NVMF_OPT_KATO,                "keep_alive_tmo=%d"     },
 605         { NVMF_OPT_HOSTNQN,             "hostnqn=%s"            },
 606         { NVMF_OPT_HOST_TRADDR,         "host_traddr=%s"        },
 607         { NVMF_OPT_HOST_ID,             "hostid=%s"             },
 608         { NVMF_OPT_DUP_CONNECT,         "duplicate_connect"     },
 609         { NVMF_OPT_ERR,                 NULL                    }
 610 };
 611
 612 static int nvmf_parse_options(struct nvmf_ctrl_options *opts,
 613                 const char *buf)
 614 {
 615         substring_t args[MAX_OPT_ARGS];
 616         char *options, *o, *p;
 617         int token, ret = 0;
 618         size_t nqnlen  = 0;
 619         int ctrl_loss_tmo = NVMF_DEF_CTRL_LOSS_TMO;
 620         uuid_t hostid;
 621
 622         /* Set defaults */
 623         opts->queue_size = NVMF_DEF_QUEUE_SIZE;
 624         opts->nr_io_queues = num_online_cpus();
 625         opts->reconnect_delay = NVMF_DEF_RECONNECT_DELAY;
 626         opts->kato = NVME_DEFAULT_KATO;
 627         opts->duplicate_connect = false;
 628
 629         options = o = kstrdup(buf, GFP_KERNEL);
 630         if (!options)
 631                 return -ENOMEM;
 632
 633         uuid_gen(&hostid);
 634
 635         while ((p = strsep(&o, ",\n")) != NULL) {
 636                 if (!*p)
 637                         continue;
 638
 639                 token = match_token(p, opt_tokens, args);
 640                 opts->mask |= token;
 641                 switch (token) {
 642                 case NVMF_OPT_TRANSPORT:
 643                         p = match_strdup(args);
 644                         if (!p) {
 645                                 ret = -ENOMEM;
 646                                 goto out;
 647                         }
 648                         kfree(opts->transport);
 649                         opts->transport = p;
 650                         break;
 651                 case NVMF_OPT_NQN:
 652                         p = match_strdup(args);
 653                         if (!p) {
 654                                 ret = -ENOMEM;
 655                                 goto out;
 656                         }
 657                         kfree(opts->subsysnqn);
 658                         opts->subsysnqn = p;
 659                         nqnlen = strlen(opts->subsysnqn);
 660                         if (nqnlen >= NVMF_NQN_SIZE) {
 661                                 pr_err("%s needs to be < %d bytes\n",
 662                                         opts->subsysnqn, NVMF_NQN_SIZE);
 663                                 ret = -EINVAL;
 664                                 goto out;
 665                         }
 666                         opts->discovery_nqn =
 667                                 !(strcmp(opts->subsysnqn,
 668                                          NVME_DISC_SUBSYS_NAME));
 669                         break;
 670                 case NVMF_OPT_TRADDR:
 671                         p = match_strdup(args);
 672                         if (!p) {
 673                                 ret = -ENOMEM;
 674                                 goto out;
 675                         }
 676                         kfree(opts->traddr);
 677                         opts->traddr = p;
 678                         break;
 679                 case NVMF_OPT_TRSVCID:
 680                         p = match_strdup(args);
 681                         if (!p) {
 682                                 ret = -ENOMEM;
 683                                 goto out;
 684                         }
 685                         kfree(opts->trsvcid);
 686                         opts->trsvcid = p;
 687                         break;
 688                 case NVMF_OPT_QUEUE_SIZE:
 689                         if (match_int(args, &token)) {
 690                                 ret = -EINVAL;
 691                                 goto out;
 692                         }
 693                         if (token < NVMF_MIN_QUEUE_SIZE ||
 694                             token > NVMF_MAX_QUEUE_SIZE) {
 695                                 pr_err("Invalid queue_size %d\n", token);
 696                                 ret = -EINVAL;
 697                                 goto out;
 698                         }
 699                         opts->queue_size = token;
 700                         break;
 701                 case NVMF_OPT_NR_IO_QUEUES:
 702                         if (match_int(args, &token)) {
 703                                 ret = -EINVAL;
 704                                 goto out;
 705                         }
 706                         if (token <= 0) {
 707                                 pr_err("Invalid number of IOQs %d\n", token);
 708                                 ret = -EINVAL;
 709                                 goto out;
 710                         }
 711                         if (opts->discovery_nqn) {
 712                                 pr_debug("Ignoring nr_io_queues value for discovery controller\n");
 713                                 break;
 714                         }
 715
 716                         opts->nr_io_queues = min_t(unsigned int,
 717                                         num_online_cpus(), token);
 718                         break;
 719                 case NVMF_OPT_KATO:
 720                         if (match_int(args, &token)) {
 721                                 ret = -EINVAL;
 722                                 goto out;
 723                         }
 724
 725                         if (token < 0) {
 726                                 pr_err("Invalid keep_alive_tmo %d\n", token);
 727                                 ret = -EINVAL;
 728                                 goto out;
 729                         } else if (token == 0 && !opts->discovery_nqn) {
 730                                 /* Allowed for debug */
 731                                 pr_warn("keep_alive_tmo 0 won't execute keep alives!!!\n");
 732                         }
 733                         opts->kato = token;
 734
 735                         if (opts->discovery_nqn && opts->kato) {
 736                                 pr_err("Discovery controllers cannot accept KATO != 0\n");
 737                                 ret = -EINVAL;
 738                                 goto out;
 739                         }
 740
 741                         break;
 742                 case NVMF_OPT_CTRL_LOSS_TMO:
 743                         if (match_int(args, &token)) {
 744                                 ret = -EINVAL;
 745                                 goto out;
 746                         }
 747
 748                         if (token < 0)
 749                                 pr_warn("ctrl_loss_tmo < 0 will reconnect forever\n");
 750                         ctrl_loss_tmo = token;
 751                         break;
 752                 case NVMF_OPT_HOSTNQN:
 753                         if (opts->host) {
 754                                 pr_err("hostnqn already user-assigned: %s\n",
 755                                        opts->host->nqn);
 756                                 ret = -EADDRINUSE;
 757                                 goto out;
 758                         }
 759                         p = match_strdup(args);
 760                         if (!p) {
 761                                 ret = -ENOMEM;
 762                                 goto out;
 763                         }
 764                         nqnlen = strlen(p);
 765                         if (nqnlen >= NVMF_NQN_SIZE) {
 766                                 pr_err("%s needs to be < %d bytes\n",
 767                                         p, NVMF_NQN_SIZE);
 768                                 kfree(p);
 769                                 ret = -EINVAL;
 770                                 goto out;
 771                         }
 772                         nvmf_host_put(opts->host);
 773                         opts->host = nvmf_host_add(p);
 774                         kfree(p);
 775                         if (!opts->host) {
 776                                 ret = -ENOMEM;
 777                                 goto out;
 778                         }
 779                         break;
 780                 case NVMF_OPT_RECONNECT_DELAY:
 781                         if (match_int(args, &token)) {
 782                                 ret = -EINVAL;
 783                                 goto out;
 784                         }
 785                         if (token <= 0) {
 786                                 pr_err("Invalid reconnect_delay %d\n", token);
 787                                 ret = -EINVAL;
 788                                 goto out;
 789                         }
 790                         opts->reconnect_delay = token;
 791                         break;
 792                 case NVMF_OPT_HOST_TRADDR:
 793                         p = match_strdup(args);
 794                         if (!p) {
 795                                 ret = -ENOMEM;
 796                                 goto out;
 797                         }
 798                         kfree(opts->host_traddr);
 799                         opts->host_traddr = p;
 800                         break;
 801                 case NVMF_OPT_HOST_ID:
 802                         p = match_strdup(args);
 803                         if (!p) {
 804                                 ret = -ENOMEM;
 805                                 goto out;
 806                         }
 807                         ret = uuid_parse(p, &hostid);
 808                         if (ret) {
 809                                 pr_err("Invalid hostid %s\n", p);
 810                                 ret = -EINVAL;
 811                                 kfree(p);
 812                                 goto out;
 813                         }
 814                         kfree(p);
 815                         break;
 816                 case NVMF_OPT_DUP_CONNECT:
 817                         opts->duplicate_connect = true;
 818                         break;
 819                 default:
 820                         pr_warn("unknown parameter or missing value '%s' in ctrl creation request\n",
 821                                 p);
 822                         ret = -EINVAL;
 823                         goto out;
 824                 }
 825         }
 826
 827         if (opts->discovery_nqn) {
 828                 opts->kato = 0;
 829                 opts->nr_io_queues = 0;
 830                 opts->duplicate_connect = true;
 831         }
 832         if (ctrl_loss_tmo < 0)
 833                 opts->max_reconnects = -1;
 834         else
 835                 opts->max_reconnects = DIV_ROUND_UP(ctrl_loss_tmo,
 836                                                 opts->reconnect_delay);
 837
 838         if (!opts->host) {
 839                 kref_get(&nvmf_default_host->ref);
 840                 opts->host = nvmf_default_host;
 841         }
 842
 843         uuid_copy(&opts->host->id, &hostid);
 844
 845 out:
 846         kfree(options);
 847         return ret;
 848 }
 849
 850 static int nvmf_check_required_opts(struct nvmf_ctrl_options *opts,
 851                 unsigned int required_opts)
 852 {
 853         if ((opts->mask & required_opts) != required_opts) {
 854                 int i;
 855
 856                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
 857                         if ((opt_tokens[i].token & required_opts) &&
 858                             !(opt_tokens[i].token & opts->mask)) {
 859                                 pr_warn("missing parameter '%s'\n",
 860                                         opt_tokens[i].pattern);
 861                         }
 862                 }
 863
 864                 return -EINVAL;
 865         }
 866
 867         return 0;
 868 }
 869
 870 static int nvmf_check_allowed_opts(struct nvmf_ctrl_options *opts,
 871                 unsigned int allowed_opts)
 872 {
 873         if (opts->mask & ~allowed_opts) {
 874                 int i;
 875
 876                 for (i = 0; i < ARRAY_SIZE(opt_tokens); i++) {
 877                         if ((opt_tokens[i].token & opts->mask) &&
 878                             (opt_tokens[i].token & ~allowed_opts)) {
 879                                 pr_warn("invalid parameter '%s'\n",
 880                                         opt_tokens[i].pattern);
 881                         }
 882                 }
 883
 884                 return -EINVAL;
 885         }
 886
 887         return 0;
 888 }
 889
 890 void nvmf_free_options(struct nvmf_ctrl_options *opts)
 891 {
 892         nvmf_host_put(opts->host);
 893         kfree(opts->transport);
 894         kfree(opts->traddr);
 895         kfree(opts->trsvcid);
 896         kfree(opts->subsysnqn);
 897         kfree(opts->host_traddr);
 898         kfree(opts);
 899 }
 900 EXPORT_SYMBOL_GPL(nvmf_free_options);
 901
 902 #define NVMF_REQUIRED_OPTS      (NVMF_OPT_TRANSPORT | NVMF_OPT_NQN)
 903 #define NVMF_ALLOWED_OPTS       (NVMF_OPT_QUEUE_SIZE | NVMF_OPT_NR_IO_QUEUES | \
 904                                  NVMF_OPT_KATO | NVMF_OPT_HOSTNQN | \
 905                                  NVMF_OPT_HOST_ID | NVMF_OPT_DUP_CONNECT)
 906
 907 static struct nvme_ctrl *
 908 nvmf_create_ctrl(struct device *dev, const char *buf, size_t count)
 909 {
 910         struct nvmf_ctrl_options *opts;
 911         struct nvmf_transport_ops *ops;
 912         struct nvme_ctrl *ctrl;
 913         int ret;
 914
 915         opts = kzalloc(sizeof(*opts), GFP_KERNEL);
 916         if (!opts)
 917                 return ERR_PTR(-ENOMEM);
 918
 919         ret = nvmf_parse_options(opts, buf);
 920         if (ret)
 921                 goto out_free_opts;
 922
 923
 924         request_module("nvme-%s", opts->transport);
 925
 926         /*
 927          * Check the generic options first as we need a valid transport for
 928          * the lookup below.  Then clear the generic flags so that transport
 929          * drivers don't have to care about them.
 930          */
 931         ret = nvmf_check_required_opts(opts, NVMF_REQUIRED_OPTS);
 932         if (ret)
 933                 goto out_free_opts;
 934         opts->mask &= ~NVMF_REQUIRED_OPTS;
 935
 936         down_read(&nvmf_transports_rwsem);
 937         ops = nvmf_lookup_transport(opts);
 938         if (!ops) {
 939                 pr_info("no handler found for transport %s.\n",
 940                         opts->transport);
 941                 ret = -EINVAL;
 942                 goto out_unlock;
 943         }
 944
 945         if (!try_module_get(ops->module)) {
 946                 ret = -EBUSY;
 947                 goto out_unlock;
 948         }
 949         up_read(&nvmf_transports_rwsem);
 950
 951         ret = nvmf_check_required_opts(opts, ops->required_opts);
 952         if (ret)
 953                 goto out_module_put;
 954         ret = nvmf_check_allowed_opts(opts, NVMF_ALLOWED_OPTS |
 955                                 ops->allowed_opts | ops->required_opts);
 956         if (ret)
 957                 goto out_module_put;
 958
 959         ctrl = ops->create_ctrl(dev, opts);
 960         if (IS_ERR(ctrl)) {
 961                 ret = PTR_ERR(ctrl);
 962                 goto out_module_put;
 963         }
 964
 965         module_put(ops->module);
 966         return ctrl;
 967
 968 out_module_put:
 969         module_put(ops->module);
 970         goto out_free_opts;
 971 out_unlock:
 972         up_read(&nvmf_transports_rwsem);
 973 out_free_opts:
 974         nvmf_free_options(opts);
 975         return ERR_PTR(ret);
 976 }
 977
 978 static struct class *nvmf_class;
 979 static struct device *nvmf_device;
 980 static DEFINE_MUTEX(nvmf_dev_mutex);
 981
 982 static ssize_t nvmf_dev_write(struct file *file, const char __user *ubuf,
 983                 size_t count, loff_t *pos)
 984 {
 985         struct seq_file *seq_file = file->private_data;
 986         struct nvme_ctrl *ctrl;
 987         const char *buf;
 988         int ret = 0;
 989
 990         if (count > PAGE_SIZE)
 991                 return -ENOMEM;
 992
 993         buf = memdup_user_nul(ubuf, count);
 994         if (IS_ERR(buf))
 995                 return PTR_ERR(buf);
 996
 997         mutex_lock(&nvmf_dev_mutex);
 998         if (seq_file->private) {
 999                 ret = -EINVAL;
1000                 goto out_unlock;
1001         }
1002
1003         ctrl = nvmf_create_ctrl(nvmf_device, buf, count);
1004         if (IS_ERR(ctrl)) {
1005                 ret = PTR_ERR(ctrl);
1006                 goto out_unlock;
1007         }
1008
1009         seq_file->private = ctrl;
1010
1011 out_unlock:
1012         mutex_unlock(&nvmf_dev_mutex);
1013         kfree(buf);
1014         return ret ? ret : count;
1015 }
1016
1017 static int nvmf_dev_show(struct seq_file *seq_file, void *private)
1018 {
1019         struct nvme_ctrl *ctrl;
1020         int ret = 0;
1021
1022         mutex_lock(&nvmf_dev_mutex);
1023         ctrl = seq_file->private;
1024         if (!ctrl) {
1025                 ret = -EINVAL;
1026                 goto out_unlock;
1027         }
1028
1029         seq_printf(seq_file, "instance=%d,cntlid=%d\n",
1030                         ctrl->instance, ctrl->cntlid);
1031
1032 out_unlock:
1033         mutex_unlock(&nvmf_dev_mutex);
1034         return ret;
1035 }
1036
1037 static int nvmf_dev_open(struct inode *inode, struct file *file)
1038 {
1039         /*
1040          * The miscdevice code initializes file->private_data, but doesn't
1041          * make use of it later.
1042          */
1043         file->private_data = NULL;
1044         return single_open(file, nvmf_dev_show, NULL);
1045 }
1046
1047 static int nvmf_dev_release(struct inode *inode, struct file *file)
1048 {
1049         struct seq_file *seq_file = file->private_data;
1050         struct nvme_ctrl *ctrl = seq_file->private;
1051
1052         if (ctrl)
1053                 nvme_put_ctrl(ctrl);
1054         return single_release(inode, file);
1055 }
1056
1057 static const struct file_operations nvmf_dev_fops = {
1058         .owner          = THIS_MODULE,
1059         .write          = nvmf_dev_write,
1060         .read           = seq_read,
1061         .open           = nvmf_dev_open,
1062         .release        = nvmf_dev_release,
1063 };
1064
1065 static struct miscdevice nvmf_misc = {
1066         .minor          = MISC_DYNAMIC_MINOR,
1067         .name           = "nvme-fabrics",
1068         .fops           = &nvmf_dev_fops,
1069 };
1070
1071 static int __init nvmf_init(void)
1072 {
1073         int ret;
1074
1075         nvmf_default_host = nvmf_host_default();
1076         if (!nvmf_default_host)
1077                 return -ENOMEM;
1078
1079         nvmf_class = class_create(THIS_MODULE, "nvme-fabrics");
1080         if (IS_ERR(nvmf_class)) {
1081                 pr_err("couldn't register class nvme-fabrics\n");
1082                 ret = PTR_ERR(nvmf_class);
1083                 goto out_free_host;
1084         }
1085
1086         nvmf_device =
1087                 device_create(nvmf_class, NULL, MKDEV(0, 0), NULL, "ctl");
1088         if (IS_ERR(nvmf_device)) {
1089                 pr_err("couldn't create nvme-fabris device!\n");
1090                 ret = PTR_ERR(nvmf_device);
1091                 goto out_destroy_class;
1092         }
1093
1094         ret = misc_register(&nvmf_misc);
1095         if (ret) {
1096                 pr_err("couldn't register misc device: %d\n", ret);
1097                 goto out_destroy_device;
1098         }
1099
1100         return 0;
1101
1102 out_destroy_device:
1103         device_destroy(nvmf_class, MKDEV(0, 0));
1104 out_destroy_class:
1105         class_destroy(nvmf_class);
1106 out_free_host:
1107         nvmf_host_put(nvmf_default_host);
1108         return ret;
1109 }
1110
1111 static void __exit nvmf_exit(void)
1112 {
1113         misc_deregister(&nvmf_misc);
1114         device_destroy(nvmf_class, MKDEV(0, 0));
1115         class_destroy(nvmf_class);
1116         nvmf_host_put(nvmf_default_host);
1117
1118         BUILD_BUG_ON(sizeof(struct nvmf_connect_command) != 64);
1119         BUILD_BUG_ON(sizeof(struct nvmf_property_get_command) != 64);
1120         BUILD_BUG_ON(sizeof(struct nvmf_property_set_command) != 64);
1121         BUILD_BUG_ON(sizeof(struct nvmf_connect_data) != 1024);
1122 }
1123
1124 MODULE_LICENSE("GPL v2");
1125
1126 module_init(nvmf_init);
1127 module_exit(nvmf_exit);